Crop Science 4: 503-508 (1964)
Implications of Genotype-Environmental Interactions in Applied Plant Breeding1
R. W. Allard and A. D. Bradshaw2
of the Department of Agronomy, University of California, Davis. Invitational
paper delivered at the Annual Meeting of the Crop Science Society of America,
Denver, Cob., Nov. 20, 1963. Some of the results reported are from work
supported by grants from the National Science Foundation (G14991) and the
National Institutes of Health (GM10476). Received Mar. 26, 1964.
2Professor of Agronomy, University of California, Davis, Calif., and Lecturer, Department of Agricultural Botany, University College of North Wales, Bangor, G. B., and Research Fellow of the Leverhulme Trustee, London. G. B.
THERE is rather general agreement amongst plant breeders that interactions between genotype and environment have an important bearing on the breeding of better varieties. However, it is much more difficult to find agreement as to what we ought to know about genotype-environment interactions and what we should do about them. Some breeders emphasize the camouflaging effect of such interactions on the value of genotypes. Consequently they attempt to estimate the magnitude of variances attributable to interactions and to utilize such estimates in developing ever more precise methods of selection. Other breeders feel that improvements in efficiency are unlikely so long as only "final" characters such as yield and quality are considered. They believe that real progress will be possible only as we clarify the pathways by which final characters are reached. Still others maintain that what is needed is a direct and pragmatic approach which will tell us what types of genetic systems are most likely to give high and stable performance.
The literature on genotype-environmental interactions is very large. It varies from reports of variety trials in the field to consideration of the mechanisms which allow cells containing the same genes to become as different as a pollen tube and an egg cell. The technologies employed are as different as those involved in the biochemical analysis of differentiation and those relevant to the development of mathematical models for population change. Probably no one has the competence to review this literature in its entirety and it is therefore important to establish at the outset the scope and limitations of a discussion of genotype-environmental interactions. The present discussion will be purposely restricted. It will attempt to summarize what is known at present that may be helpful to the practising plant breeder in developing varieties which minimize unfavorable genotype-environmental interactions, i.e. varieties which are able to control their developmental processes in such a way as to give high and consistent performance.
Classification of Genotype-Environmental Interactions
|Figure 1. Graphical representation of some genotype-environment interactions.|
It is important at the very outset to classify the outward manifestations of interactions between genotype and environment. The problem of classification is complex but, as was pointed out by Haldane (10), certain facts about it are so simple they are usually not stated. Assume 2 genetically different populations, A and B, and 2 environments. X and Y. Further assume that measurements are made on some character (say yield) and that significant differences are obtained such that the 4 genotype-environment combinations can be placed in rank order 1 to 4. Twenty-four interaction types are possible among which 6 are shown in Figure 1. Some of the points to note about these particular interaction types is whether genotype A does better than B in each environment (as in type 1), whether A is superior to B in one environment and inferior to the other (as in type 4), and whether the change in environment affects the 2 genotypes in opposite directions (as in type 3). Each of these interactions represents a type well known to plant breeders. For example, in types 1 and 3. genotype A can be taken as the universal variety; in type 6, A and X can be taken as specialized variety and matching favorable environment and B and Y as a less specialized variety and less favorable environment; with type 4, both A and B are specialized varieties and X and Y are specialized environments.
In practice the situation is immensely more complicated because many genotypes and environments must be considered and the number of possible types of interaction is very great. If there are only 2 genotypes and 3 environments, and a single criterion of classification (say yield), 60 types of interaction are possible. For m=10 genotypes and n=10 environments there are (mn)!/m!n!=10145 possible types of interaction, which is larger than the total number of plants which have ever existed on earth. Hence for even modest numbers of genotypes, environments, and criteria of classification the number of interaction types is so large there is no advantage in attempting to generalize the classification.
There are, however, three points which are worth making. First, even though only a small proportion of the possible number of interactions may have any importance for the breeder, the chance that even these can be analyzed and explained in terms of basic causes is very small. Second, estimates of the magnitude of genotype-environmental interactions, which must necessarily be made from very small samples relative to the whole, are likely to provide little more than gross approximations of the total potential of such interactions. Third, and perhaps most important, the above classification shows that among the virtual infinity of possible interactions, there is only one in which a single genotype is best in all environments. It is unlikely that such a genotype exists even in species which occupy relatively limited and uniform environments. Nevertheless, improvement has been made in productivity in virtually all crop plants: this indicates that breeders have overcome some of the hurdles that the complexity of genotype-environment interactions pose to progress and it holds out hope for further gains. Let us now attempt to analyze past progress for clues to more efficient future progress.
In the past the attention of practical plant breeders has centered on "final" characters. However, plant breeders are fully aware that higher plants are dynamic living systems in which change occurs constantly from germination to maturity. The pattern of change is rarely the same from genotype to genotype in one environment or for a single genotype grown in different environments. It has been almost an article of faith from the earliest days of plant breeding that, if we only understood the developmental pathways by which final characters are reached, this would help us to improve the efficiency of breeding.
Interest in this topic has received great stimulus recently with the elucidation of the genetic information code. The code tells us that differences between genotypes, while based on a small number of elements, arise in the almost infinite variety of molecules which can be formed of these few elements, and that there are endless numbers of ways in which these molecules can be organized at subcellular, at cellular, and at various higher levels of organization. In highly organized systems, such as those of the angiosperms, the sequences in which events take place are certainly so intricate as to allow a large number of opportunities for fluctuations in environment to induce interactions at each of the various levels of organization. While this provides insight into the complex causes of genotype-environmental interactions and opens new and more soundly based avenues by which they can be investigated, many basic problems at the molecular level remain unsolved. Among these the most important center about the regulation of gene action and, until this problem is solved, it is unlikely that we can understand how genes act to provide the diversity of developmental patterns which underlie genotype-environmental interactions. In other words, the gap between genetic code and final character remains very large. However, it is encouraging that an increasing number of attempts are being made to consider the pathways above the molecular level through which final characters are reached. These attempts vary from biochemical studies of enzyme systems, through studies of growth patterns, to analyses based on separating final characters into components. A few examples will suffice.
An example of the pathway approach at the enzyme level is provided by the study of Sarkissian and Huffaker (27) on the effect of chloramphenicol on carboxylation systems in isogenic lines in barley. The importance of CO2 fixation in plant growth is obvious and, interestingly, the heterozygote at the hooded locus was superior to the homozygotes in carboxylation and in growth rate when treated with chloramphenicol. It is not clear from this experiment whether the superior activity and greater stability of the carboxylation apparatus in heterozygotes is a cause or manifestation of heterosis, nor is it known whether the heterozygote is superior and more consistent in yield than the homozygotes under field conditions. However, these uncertainties in themselves illustrate an important point. Any component or pathway that can be measured with less work and error than a final character, and which is highly correlated with the final character under field conditions, will obviously be useful as an aid to plant improvement. For such a purpose the cause-effect relationship is immaterial.
The study of Schwartz (28) on the genetic control of an enzyme with esterase activity is another example of the pathway approach at the biochemical level. In this study evidence was obtained for the formation of "hybrid" enzymes in heterozygotes. The discovery of "hybrid substances" is especially interesting because they could be a factor in the better balanced genetic systems of hybrid varieties.
An example of another level is provided by the work of Langridge and Griffing (17) and Langridge (9) on Arabidopsis thaliana. They found that hybrids exhibited greater mean growth than parents, especially at high temperatures, and greater stability of phenotypic expression over a wide temperature range. They presented evidence that homozygotes cease growth because of deficiencies in particular enzymes which distinguish different homozygous ecological races and they postulated that vigor and stability are a consequence of the combination in the hybrid of alleles which condition more thermostable products.
A great deal of work has been recently carried out at the other end of the pathways and lead to final manifestation of characters. The work of Watson (33), Thorne (32), and others has encouraged investigators to envisage final yield as the result of the integrated activity of various contributing organs-leaves, parts of the stem, and inflorescence. Some breeders, notably Williams (34) and Grafius (8), have taken the position that studies of individual yield components (i.e., somatic analysis) can lead to simplification in genetic explanation and hence are valuable to breeders in prediction and determination of the effects of the environment. Their viewpoint has received support from the work of Duarte and Adams (5) who have shown that heterosis in leaf area in beans could be explained by interaction in terms of leaflet number and size, not at the genic but at the component level.
However, it is likely that not all interactions can be so simply explained and some investigators (e.g., Hayman (11), Moll et al. (22)) argue that, while such components may be one step closer to the primary effects of genes, they are nevertheless still many steps removed; they argue further that the components may be extraneous to yield and hence that yield may be closer to the primary effect of genes than the expression of any of its components. Leng (19) found that fluctuations in environment led to striking interactions between components and concluded that analyses of phenotypic components did not provide a satisfactory basis for appraisal of yield.
Aside from the satisfaction that would accompany understanding of the biochemical, physiological, or morphological basis of the interplay between genotype and environment, the question arises whether studies of basic causes have anything to offer the practising breeder whose primary responsibility is to develop and identify superior varieties.
It is possible that poor performance in certain environments results from failure of some single system or a few systems to fulfill requirements in the sequence of events despite continuing operation of other systems at nonlimiting rates. Ketellaper (16) has made extensive surveys of such cases and has succeeded in curing some such "climatic lesions" chemically. Examples are cures of low temperature effects in eggplant by a mixture of ribosides, of high temperature effects in peas and broad beans by ascorbic acid, and in a native California lupine by application of a vitamin B mixture. Such information about the basis for unfavorable genotype-environment interactions should be helpful to the breeder in developing "genetic cures". However, such ideas are still a long way from realization and the fact remains that, despite promise for the future, studies of basic causes have as yet contributed little that is directly useful to the breeder. To determine what may be profitable in terms of present day technology we therefore have to turn to analyses of genotype-environment interactions as we actually find them.
Significance of Genotype-Environmental Interactions
Variations of the environment can be divided into two sorts: predictable and unpredictable. The first category includes all permanent characters of the environment, such as general features of the climate and soil type, as well as those characteristics of the environment which fluctuate in a systematic manner, such as day length. It also includes those aspects of environment that are determined by man and can therefore be fixed more or less at will, such as planting date, sowing density, methods of harvest, and other agronomic practices.
The second category includes fluctuations in weather, such as amount and distribution of rainfall and temperature, and other factors, such as established density of the crop. In an inefficient agricultural system it may also include variations in agronomic practice which in more advanced agricultures might be held reasonably constant.
The distinction between these two categories is not always clear cut, and the characteristics included will vary from crop to crop. Nevertheless, the various qualifications which can be applied should not be allowed to mask the essential difference between the two categories. This is because they have distinctly different impacts, not only on operational procedures at the selection stages of breeding programs, but also in testing stages.
Predictable environmental variation. With predictable variation the essential first step is to recognize the existence of differences.
Large environmental differences, such as the difference between oceanic and continental climate, present no problem, but differences which are small or difficult to measure without elaborate tests or apparatus may be troublesome. Usually the crop itself will be the best indicator of the importance of these predictable variations as modern techniques of plant analysis for nutrient deficiencies have shown. Large variety × location interactions when the crop is tested throughout a region indicate that the region includes a number of different and special environments. Similarly, large variety × treatment interactions, such as interactions between genotypes and fertility levels, sowing dates, and so forth, indicate that the treatments induce special environments.
Significant variety × location or variety × treatment interactions suggest immediately that the appropriate breeding program should allow for the development of a number of varieties each particularly adapted to one of the special environments. Such a course of action is usually feasible because there seems to be no limit to the variability available enabling plants to adapt to specific conditions of temperature, photoperiod, soil fertility, method of harvesting, and the like. In certain cases the environmental factor may be one with adverse effects, which could be remedied if suitable agronomic or other steps were taken, e.g., correction of salinity. However, it may often be easier to alter the genotype of the crop instead, i.e., to cure the genotype rather than the environment, as discussed by Epstein (6).
A further aspect of this problem may seem obvious to breeders of some crops, but it has caused much confusion in the past in others. This is the need to assess breeding material under conditions as similar as possible to the conditions in which it will eventually be used: this is often not done. In cereals, for example, it is usually not easy to test initial materials at the spacings used in commercial practice. In grasses and other herbage crops the problem is extremely complex and practical considerations usually make it necessary to assess the material as single spaced plants even though the material will ultimately be grown in closely knit swards. Lazenby and Rogers (18) have recently shown the pitfalls of this in Lolium perenne, where the rank of four varieties was radically altered when the material was tested as spaced plants and in broadcast swards. As a result, techniques of assessment need to be appraised critically and in many cases new techniques developed.
Unpredictable environmental variation. The implications of variety × year interactions are very different from variety × location or variety × treatment interactions. This is because year-to-year fluctuations cannot be predicted in advance and the breeder can hardly aim his program at developing varieties suited to special circumstances he cannot foresee. In varietal trials it is common to find large variety × year and large variety × year × location interactions. As an example, Rasmussen and Lambert (25) found the variety × year component was more than 4 times as large as the variety × location component in barley trials over 4 years at 8 locations in Minnesota. The variety × location × year component was larger still. Violent year fluctuations in selective values have recently been demonstrated in lima beans (2).
Such findings have an important bearing on methods of testing varietal differences. Within a region over which it is likely that a set of varieties will be adapted, it is essential that tests be conducted in a series of locations over a series of years. Great precision in the conduct of any one trial at any one location is unnecessary and may be wasteful. In an assessment of spring oat varieties carried out in Great Britain, it was concluded that 1 replication was adequate, and 2 replications quite sufficient in any 1 place in any year (26). These findings also have a bearing on testing in the development stages of breeding programs, during which the breeder attempts to select superior genotypes from genetically variable populations. If testing is carried out at a number of places, chances of identifying genotypes adapted to several environments are improved. At the same time difficulties of testing selection materials at many locations are formidable. The development of efficient methods for testing genetically variable populations under a range of environmental conditions represents a problem of great enough importance to justify careful thought and extensive experimentation.
But there are some further, and perhaps more important, implications of unpredictable environmental fluctuations regarding breeding goals. If it can be argued that varieties should be developed with specific adaptation to predictable special environments, then it should also be a goal of plant breeding to produce varieties that are adapted to withstand unpredictable transient environmental variations. These will be well buffered varieties that are able to adjust their life processes in ways such as to maintain productivity at a high level despite unpredictable fluctuations of the environment. However, this raises the question whether such stability is under genetic control, and whether certain types of genetic systems are more likely to lead to stability in productivity than other genetic systems. This is a key question in applied breeding which deserves examination in some detail.
Mechanisms Promoting Stability in Productivity
It is important to emphasize that the stability with which we are concerned does not imply general constancy of phenotype in varying environments. It implies stability in those aspects of phenotype, especially yield and quality, that are important economically. Such stability may in fact depend on holding some aspects of morphology and physiology in steady state and allowing others to vary. Thus the required varieties will show low genotype-environment interaction for agriculturally important characters, particularly yield, but not necessarily for other characters. A variety which can adjust its genotypic or phenotypic state in response to transient fluctuations in environment in such ways that it gives high and stable economic return for the place and year can be termed "well-buffered". This term is therefore equivalent to "homeostatic", used in the sense of Lewontin (21), but the latter term will not be used owing to controversy over its meaning in the literature.
There are two obvious general ways in which a variety can achieve stability. First, the variety can be made up of a number of genotypes each adapted to a somewhat different range of environments. Second, the individuals themselves may be well buffered so that each member of the population is well adapted to a range of environments. Genetically homogenous populations, such as pure line varieties or single crosses, obviously depend heavily on "individual buffering" to stabilize productivity, whereas both paths are open to genetically heterogeneous populations. The terms "individual buffering" and "populational buffering" are adopted here to describe these two methods of stabilization of yield because these terms are noncommital as to the mechanisms which may produce the stability.
Individual buffering. In outbreeding species there is a good deal of work which indicates that buffering is conspicuously a property of a heterozygotes. In a general summary of work with animals Lerner (20) reached the conclusion that "adaptedness, the attribute of individuals to be fit in the Darwinian sense to their immediate environment, is mediated by heterozygous advantage in buffering ability." The situation seems much the same in outbreeding plants. The experiment of Shank and Adams (29) in which they compared inbreds of maize with hybrids with respect to six characters provides a good example. Coefficients of variability were larger for the inbreds than for the hybrids and at the same time varied markedly between inbreds. The differences among the inbreds show that buffering is a feature of the specific genotype while the greater variability of inbreds as a group, compared to hybrids as a group, shows that buffering is also a feature of heterozygosity.
Further evidence comes from the recent work of Clausen and Hiesey and their associates. They compared the growth of individual races of a number of species at three altitudes in California, together with F1 and F2 hybrids between certain races. They found in Mimulus (12) and in Potentilla glandulosa (3) that the parental races themselves cannot survive at all three altitudes but that F1 hybrids can survive and are as vigorous as each parent in its optimal environment. In Mimulus neither parent can survive in anything other than its own environment, while the F1 hybrid is successful in all three environments. In the F2 generation some individuals can be found which equal or even surpass the F1 hybrids. In these cases the hybrids are remarkably well buffered because the three altitudinal stations have climates which range from Mediterranean to subalpine.
In inbreeding species there is evidence that buffering can be a property of specific genotypes not associated with heterozygosity. Cereal breeders, for example, have considerable practical experience to indicate that there are varietal differences in degree of buffering. An example in barley is the comparison between Atlas and Vaughn. Atlas is widely distributed throughout California and yields satisfactorily in very contrasting seasons. Vaughn, although superior in yield to Atlas under optimal cultural conditions, is otherwise an erratic producer. Williams (35) and Jinks and Mather (15) have shown differences in the stability of homozygous lines and absence of increased stability in F1 hybrids in tomatoes and in Nicotiana rustica, respectively.
Unfortunately, the role that heterozygosity plays in self-pollinated species has been made obscure by failure on the part of some investigators to distinguish between general constancy of phenotype and stability in fitness characters. Some good evidence regarding fitness characters comes from the work of Griffing and Langridge, mentioned earlier, on lower genotype-environment interaction of heterozygotes in Arabidopsis as a result of their greater thermostability. At the same time they demonstrated the additive variability also existed for this character indicating the possibility of improvement by selection.
Additional evidence on the particular stability of heterozygotes comes from studies of wheat and beans. When Palmer (23) grew homozygous lines of wheat and their F1 hybrids under field conditions he found markedly reduced variance in the seed production of F1 hybrids. Observations over many years in lima beans indicate that homozygous lines and their F1's make similar yields in years when seed yield is high. But in a poor year F1's often yield half again as much as their homozygous parents and sometimes twice as much. Still further evidence in lima beans comes from estimates of selective values of the two homozygotes and the heterozygote for marker genes under population conditions (2). In studies conducted over 10 years it was found that homozygotes and heterozygotes tended to contribute more or less equal numbers of progeny to the next generation in years when seed yields were high. But in poor years the heterozygotes sometimes contributed more than twice as many offspring to the next generation as the homozygotes. These results thus tend in the same direction as those of Griffing and Langridge: under optimal conditions homo. zygotes and heterozygotes differ little in fitness but as conditions deviate more and more widely from optimum the advantages of heterozygotes increases progressively.
Populational buffering. Populational buffering refers to buffering above and beyond that of individual constituents of populations, i.e., buffering which arises in interactions among different coexisting genotypes. Like individual buffering it should be measurable in terms of genotype-environmental interactions. The most precise information on populational buffering comes from comparisons between pure line varieties grown singly and in mixture. Simmonds (30) recently reviewed the literature on this topic and found that mixed populations are nearly always more stable in yield than their components. In wheat, for example, coefficients of variability over seasons were about two-thirds as large for mixtures (7.3%) as for homogeneous populations (11.6%). There was some suggestion that the stabilizing effect was much greater for some combinations than for others. In mean yielding ability the average advantage of mixtures over the means of components was of the order of 3 to 5% but many mixtures outyielded the higher component when tested over several years. Since the mixtures were compounded more or less randomly from good local varieties, without benefit of information about combining ability, it seems possible the stability which often accompanies genetic diversity may have substantial contributions to make in improving crop yields.
Information on populational buffering in heterozygous materials comes primarily from comparisons of single and double crosses in corn. Jones (14) analyzed extensive yield trials and found that coefficients of variability were smaller for double crosses (12.3%) than for single crosses (21.4%). Sprague and Federer (31) found that variety × location and variety × year interactions were smaller for double crosses than for single crosses, which also suggests greater stability of double crosses. Jones attributed this stability to the buffering effect of heterogeneity and suggested that it is stability that allows double crosses to make high mean yields over many years, even though the highest yield in any one place and year is likely to be obtained from some particular single cross.
The work of Finlay (7) on barley and Allard (1) on lima beans shows that advanced generation hybrid populations in self-pollinated species are often highly buffered. Finlay found that 45 F2 barley hybrids outyielded their parents substantially in the variable environment of South Australia and that they were also markedly superior in stability of productivity. Finlay emphasized that much of the advantage of the F2 populations resulted from striking superiority in stress environments. In lima beans Allard found that 3 unselected F7 populations outyielded their parents by 7%, as an average, over 16 environments. The F7 populations achieved superiority through steady good performance whereas the parental varieties tended to be efficient in some environments and inefficient in others. It is not possible to assess the extent of populational buffering in the above studies of barley and lima beans because the populations were heterozygous to some extent as well as heterogeneous.
Much of the evidence on populational buffering is not critical. Nevertheless, it is cumulatively very powerful.
CONCLUSIONS AND COMMENTS
We are led by the evidence reviewed to the following conclusions. First, interactions containing genotype × year terms are particularly interesting to applied plant breeders. This is because they reflect fluctuations in environment which for the most part cannot be predicted in advance and hence can be countered only by developing varieties in which developmental sequences are canalized along pathways that lead to high performance. Second, in the search for such varieties it is important to learn more about the basic causes of interactions. However, for the present, developmental genetics has relatively little to offer that is useful at the operational level. Finally, we conclude that genetic diversity, either in heterozygotes or in mixtures of different genotypes often leads to stability under varying environmental conditions.
One aspect of genetic diversity, i.e., the diversity associated with heterozygosity, has been widely recognized and utilized in outbreeding species. More recent evidence, including extensive commercial experience with F1 hybrid varieties in grain sorghums (24) suggests that heterosis, and the individual buffering commonly associated with it, may also have substantial contributions to make in increasing and stabilizing yield in self-pollinated species. The other aspect of genetic diversity, i.e., populational buffering, is much less widely recognized and there have been few conscious attempts to exploit its possible advantages in reducing unfavorable genotype-environmental interactions. There seems to be little doubt that populational buffering is also not only real, but is often important, even though little is known of its underlying mechanisms. The question is whether it can be put to practical use.
The answer to this question depends in part on the agricultural properties of various types of populations which might exploit such heterogeneity. In some cases, as for example in certain canning or freezing vegetables, the demands for uniformity of product may be so compelling as to preclude the use of heterogeneous populations. However in most crops it should be possible to develop mixed populations which satisfy basic requirements concerning uniformity of maturity, height and other plant characteristics, as well as the requirements for uniformity of product. Processors usually deal in blends anyway and it makes little difference whether the blending is done in the field or in the mill, or processing plant, so long as the characteristics of the blend are understood.
Part of the answer must also depend on the biological and economic feasibility of various sorts of populations. In corn, widespread use of populational buffering is already being made through the growing of genetically heterogeneous double-cross hybrids. However, a number of other types of populations seem feasible, including deliberately compounded mixtures of single crosses, mixtures of double crosses and synthetic varieties. In self-pollinated species some of the possibilities are mixtures of homozygotes, mixtures of single crosses (which are now possible in grain sorghums), and advanced generation hybrid populations.
It can be argued that homozygous genotypes and homogeneous populations can ultimately be produced which will cope with unpredictable fluctuations in environment as well as any other type of population. This may be the case but in the meantime heterozygous and heterogeneous populations appear to offer a greater opportunity to produce varieties which show small genotype-environment interaction through versatility and resilience.
See also: Gustafsson: Cooperation among genotypes (1953)